Using EHRs for Heart Failure Therapy Recommendation Using Multidimensional Patient Similarity Analytics

Area of Science:

• Health Informatics
• Machine Learning in Medicine
• Computational Biology

Background:

• Electronic Health Records (EHRs) offer rich patient data for understanding disease diagnosis and prognosis.
• Existing patient similarity methods often rely solely on diagnosis data from EHRs.
• A need exists for multidimensional approaches integrating various EHR data types.

Purpose of the Study:

• To develop a multidimensional patient similarity assessment technique using diverse EHR information.
• To predict individualized medication plans for new patients based on similar historical cases.
• To enhance personalized medicine through advanced data analytics.

Main Methods:

• Developed a novel algorithm integrating multiple data types from EHRs for patient similarity assessment.
• Employed hierarchical clustering to group patients based on shared characteristics.
• Assigned medication plans by calculating similarity indices against the broader patient population.

Main Results:

• Evaluated the approach on a cohort of 1,386 heart failure patients from Mayo Clinic EHR data.
• Achieved an Area Under the Curve (AUC) of 0.74, demonstrating significant predictive performance.
• Validated the feasibility of using population-based EHR data for individual patient assessment.

Conclusions:

• Multidimensional patient similarity assessment using EHR data is effective for personalized medication planning.
• Hierarchical clustering and similarity indices provide a robust framework for patient stratification.
• This approach holds promise for improving patient-specific healthcare decisions and outcomes.